Sie sind auf Seite 1von 11

440

IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS, VOL. 1, NO. 4, DECEMBER 2011

A Versatile Wireless Portable Monitoring System for BrainBehavior Approaches


Da-Wei Chang, Member, IEEE, Sheng-Fu Liang, Member, IEEE, Chung-Ping Young, Member, IEEE, Fu-Zen Shaw, Alvin W. Y. Su, You-De Liu, Yu-Lin Wang, Yi-Che Liu, Jing-Jhong Chen, and Chun-Yu Chen

AbstractIt is critical to set up a precise and feasible monitoring system for a variety of animal and human studies. A multichannel wireless system for monitoring physiological signals of freely moving rats is presented. This system combines electroencephalogram (EEG) and acceleration signals, enabling the study of association between brain and behavior. A combination of EEG and accelerometers eliminates the necessity for complicated video installation as well as time-consuming and tedious analysis of recorded videos. The IEEE 802.15.4 based wireless communication frees the experimental subject from the hassle of wires and reduces wire artifacts during recording. Long-period continuous recording was possible because of the low power feature of the system. Methods for automatic wakesleep state discrimination and temporal lobe epileptic seizure detection are also proposed to demonstrate the advantages of the system. An accuracy of up to 96.22% for the automatic discrimination of wakesleep states is an advantage of our system. In addition, the detection of amygdala-kindling temporal lobe seizures reaches 100% with zero false alarms, greatly saving manpower in the identication of temporal lobe epilepsy. Index TermsAccelerometer, electroencephalogram (EEG), epilepsy, vigilance, wireless.

I. INTRODUCTION

T IS readily accepted that the dynamics of an electroencephalogram (EEG) changes markedly with different stages of wakefulness and sleep [1][3]. To accurately ascertain the state of vigilance, an EEG is commonly used and is often incorporated with other biosignals, such as an electromyogram (EMG) or an electrooculogram (EOG). However, excess recording wires are cumbersome and frequently cause state transitions, which inuence the determination of the state. A combination of video and EEG is another popular approach in the clinic, but the video provides low spatial and temporal

Manuscript received May 31, 2011; revised October 06, 2011; accepted December 10, 2011. Date of publication January 11, 2012; date of current version February 01, 2012. This work was supported in part by the National Science Council of Taiwan under Grant NSC 100-2218-E-006-023, Grant NSC 100-2220-E-006-010, and Grant NSC100-2220-E-006-011. This paper was recommended for publication by Guest Editor J. Sanchez. D.-W. Chang, S.-F. Liang, C.-P. Young, A. W .Y. Su, Y.-D. Liu, Y.-L. Wang, Y.-C. Liu, J.-J. Chen, and C.-Y. Chen are with the Department of Computer Science and Information Engineering, National Cheng Kung University, Tainan 70101, Taiwan (e-mail: davidchang@csie.ncku.edu.tw; siang@mail. ncku.edu.tw; cpyoung@mail.ncku.edu.tw; alvinsu@mail.ncku.edu.tw; yodaliu@gmail.com; daphne.yl.wang@gmail.com; liu750815@gmail.com; yellow3899@gmail.com; cychen7554@gmail.com). F.-Z. Shaw iswith the Department of Psychology, National Cheng Kung University, Tainan 70101, Taiwan (e-mail: fzshaw@mail.ncku.edu.tw). Digital Object Identier 10.1109/JETCAS.2011.2181454

resolution for the study of the association between the brain and behavior. Moreover, video monitoring also has an ethical consideration for daily life. Several groups have proposed alternative approaches instead of video. Currently, accelerometers fabricated with microelectromechanical systems (MEMS) have a tiny size and low weight and can provide ne temporal information about behavior alterations. Using an accelerometer in the eld has been demonstrated to be successful for motion detection [4], [5] and for the discrimination of daily life activities [6]. These studies, however, provide no information about brainbehavior association. Recently, a study demonstrated that the combination of a head accelerometer with an EEG improves the accuracy of discrimination in a state of vigilance [7]. In addition to the classication of vigilance states, the identication and categorization of epileptiform activity from video-EEG recordings of several hours or days is tedious and time-consuming work that requires a signicant investment of labor; however, it is essential for treating epileptic patients in the clinic. Numerous methods, such as Lyapunov exponents [8], [9] or complexity analyses of EEGs using entropy [10][13], have been proposed as a means of marking seizure discharges. A major limitation of available methods is that errors in seizure identication may occur when a technician operates for a considerable duration. A robust and automatic system to assist in seizure identication would be benecial for shortening time consumption, lowering unexpected errors, and saving manpower. Recently, the comprehension of brain activity-driven motor activity has received a substantial amount of attention and tends to correspond to a multimillion-dollar industry, such as the application of brain computer interfaces [14], [15]. During a seizure, the EEG of epileptic patients is often characterized by high-amplitude and synchronized EEG waveforms with dynamic and versatile features. Thus, the analysis of seizure-related motor activity may be useful for understanding brainaction relationships [16]. Moreover, wireless and embedded systems have been widely applied in several aspects of physiological or EEG signal monitoring [12], [17][19]. In this paper, a wireless monitoring system that is useful for understanding brainaction relationships is proposed. This system monitors both EEG and acceleration signals of freely moving subjects. The wireless link frees the subject from the hassle of wires and reduces wire artifacts during recording. In addition, novel and useful techniques for improving the discrimination of vigilance states and for the identication of temporal lobe epilepsy (TLE), using the cooperation of EEGs and accelerometers, have also been proposed. The system was

2156-3357/$26.00 2011 IEEE

CHANG et al.: A VERSATILE WIRELESS PORTABLE MONITORING SYSTEM FOR BRAINBEHAVIOR APPROACHES

441

tested on rats to validate its performance with the conrmation of video examination. The ultimate objective of this research is to develop an implanted medical device that implements state-dependent closed-loop seizure detection and that controls for daily activities in cases where continuous video monitoring would not be feasible. The remainder of this paper is organized as follows. Section II presents the surgical and experimental procedures. Section III describes the system implementation, and Section IV presents the two techniques developed for the vigilance state discrimination and the TLE identication. System evaluation results are shown in Section V, and the results of the vigilance state discrimination and the TLE identication are presented in Sections VI and VII, respectively. Finally, a discussion and conclusions are given in Section VIII.
Fig. 1. Block diagram of the physiological signal monitoring system.

II. SURGICAL AND EXPERIMENTAL PROCEDURES Adult Wistar rats were used in this study. Animals were kept in a room under a 12:12-h light-dark cycle with food and water provided ad libitum. All surgical and experimental procedures were reviewed and approved by the Institutional Animal Care and Use Committee of National Cheng Kung University. Rats were anesthetized by sodium pentobarbital (50 mg/kg, i.p.). Subsequently, each rat was placed in a standard stereotaxic apparatus. Screw electrodes were bilaterally implanted over the frontal cortex (anterior 2.0 mm, lateral 2.0 mm with regard to the bregma). The microelectrode was made of twisted Teon-coated stainless steel microwires (#7916, A-M Systems, Carlsborg, WA) insulated except at the tip (0.5 mm); the tip separation was 0.5 mm. The microelectrodes were implanted into the bilateral basolateral amygdala (posterior 2.6 mm, lateral 4.8 mm, and ventral 8.5 mm from the bregma). A ground electrode was implanted 2 mm caudal to the lambda. All electrodes were connected to a miniature receptacle, which was attached to the skull with dental cement. After surgery, the rats were allowed two weeks of recovery. Before the experiment, each rat was placed in the recording environment at least ve times (1 h/day) prior to testing to allow rats to become familiar with the experimental apparatus, including wearing the jacket and carrying the battery. Subsequently, the right basolateral amygdala kindling was performed. The amplitude of the stimulation current used for kindling was set at 400 A, which has previously been found to be sufcient to kindle rats [20], [21]. To kindle from the right amygdala, a stimulation of 400 A, 80 Hz, 1 ms pulse trains for a total duration of 2 s were delivered every 20 min through an isolated constant current stimulator (Grass S44). Up to 20 stimulations daily were conducted for two days. Seizure stages were observed after each stimulus and were classied using Racines standard ve-stage scale [22]: stage 1, facial movement; stage 2, rhythmic head nodding; stage 3, unilateral forelimb clonus; stage 4, bilateral forelimb clonus and rearing; and stage 5, falling and tonic-clonic convulsion. Based on empirical results, 30 amygdala stimulations could elicit stage 5 on Racines scale. Up to 10-h of continuous recording was conducted in two rats after 40 amygdala stimulations, for the classication of vigilance states. Subsequently, recordings of at least 3 h with amygdala kindlings of interstimulus intervals of 2040 min were performed to assess the system performance for seizure detection. III. SYSTEM ARCHITECTURE AND IMPLEMENTATION A. Hardware Architecture The physiological signal monitoring system is composed of a data acquisition device that is carried by each experimental subject for acquiring the EEG and accelerometer signals and a nearby host computer congured as a virtual instrument (VI) for remote real-time physiological signal and behavioral state monitoring. The device and the host computer communicate with each other based on a 2.4 GHz wireless IEEE 802.15.4/ZigBee protocol. The host computer, which was originally not capable of IEEE 802.15.4 communication, utilizes a RS-232 port to connect with a Texas Instrument SmartRF04 Evaluation Board with a CC2430 Evaluation Module for EEG and accelerometer data reception [23]. Fig. 1 depicts the functional block diagram of the system. The data acquisition device consists of three modules: an accelerometer board, an EEG amplication board, and a microcontroller board. The former two boards include a three-axis accelerometer and a signal conditioning circuit for EEG data acquisition, and the latter board is responsible for data buffering and wireless transmission. Fig. 2 depicts both sides of each prototype circuit board of the data acquisition device. Each module is described below in more detail. 1) Accelerometer Board: A three-axis micromachined accelerometer (MMA7260QT, Freescale Semiconductor; congured at a sensitivity of 600 mV/g and a range of 2 g) [24] was mounted on the accelerometer board, and the output signals were fed to the analog-to-digital converter (ADC) of the CC2430 chip on the microcontroller board. 2) EEG Amplication Board: Up to four channels of EEG signals were amplied and band-pass ltered (1000x, 0.880 Hz), and then the amplied EEG was positively biased to the input voltage range of an ADC of the CC2430 chip on the microcontroller board. 3) Microcontroller Board: The core component on the microcontroller board is a CC2430 system-on-chip RF IC, which includes the IEEE 802.15.4 based CC2420 RF trans-

442

IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS, VOL. 1, NO. 4, DECEMBER 2011

Fig. 2. Both sides of the data acquisition devices: accelerometer board (left), microcontroller board (middle), and EEG amplication board (right).

TABLE I OPERATION AND SPECIFICATION OF THE DATA ACQUISITION DEVICE

Fig. 3. Flowchart of the wireless transmission and data acquisition.

ceiver, an enhanced 8051 microcontroller unit (MCU), 128 KB ash memory, 8 KB RAM, and other peripherals such as ADCs and timers [25]. The optimized 8051 core typically possesses eight-fold performance of a standard core and it has a low current consumption (12.3 mA when running at 32 MHz). The CC2420 RF allows for software-controlled activation and deactivation. When the RF is activated, its current consumption dominates the overall current consumption of the microcontroller board (e.g., 26.9 mA and 26.7 mA during data transmission and reception, respectively, with low microcontroller activity). The MCU was programmed to control the sampling period of the ADC, to buffer and transmit sampled data through the wireless link, and to control the wireless RF state. Timer 1 was congured at 500 Hz as the sampling rate of a 10-bit ADC under a 1.25 V internal reference voltage. Because of the limitation of the ash memory size, the sampled data cannot be stored on the microcontroller board for a long duration. Instead, the data are buffered and the IEEE 802.15.4 wireless communication is periodically activated to transmit the data to the host computer when the buffer is full. Table I summarizes the data acquisition operation and the MCU specication of the data acquisition device. B. Software Implementation The software of the system is divided into two parts; one executes on the host computer and the other runs on the microcontroller board. The software developed on the host computer

is used for data storage and physiological signal monitoring, while the rmware on the microcontroller board performs data acquisition and wireless communication. Each part is described below in more detail, respectively. 1) Data Acquisition: The data acquisition task controls the conversion of the conditioned EEG and the accelerometer signals to the digitally sampled data. Timer 1 interrupt is congured for a 2-ms sampling period for the start of a sequence of up to 7 A/D conversions, each of which takes 36 s for a 10-bit resolution result. Upon the completion of the conversion, the result is copied to a transmission buffer. The sampled data in the buffer are transmitted to the host computer through the wireless link when the buffer is full. 2) Wireless Transmission: To reduce the power consumption of the battery-powered data acquisition device, the CC2420 RF is activated only when the transmission buffer is full. In the current implementation, the buffer size is congured to accommodate 28 sampled data. As a consequence, the wireless communication is activated and the buffered data are transmitted to the host computer based on the IEEE 802.15.4/Zigbee protocol every four sampling periods if all of the seven ADC channels are used. The RF is turned off immediately after transmission completion. The wireless communication protocol stack and application programming interface were provided free by the chip company. Fig. 3 shows the owchart of the rmware, including the main program for wireless RF control and data transmission (left) and the interrupt service routine (ISR) for Timer 1 (right). 3) Backend Host Computer: A PC is used as a central console for an online monitoring experimental subject in an animal laboratory and as a data server for remote ofine study of previous experimental results. The required storage space on a PC is 119 MB/h or 2856 MB per day, based on four EEG and three accelerometer channels, a 500-Hz sampling rate and a 10-bit ADC resolution. The real-time online physiological signal monitoring was developed using National Instrument LabVIEW to create a graphical user interface (GUI) VI, to monitor and store the sampled data.

CHANG et al.: A VERSATILE WIRELESS PORTABLE MONITORING SYSTEM FOR BRAINBEHAVIOR APPROACHES

443

four state indices corresponding to wake, nonrapid eye movement (NREM) sleep, rapid eye movement (REM) sleep, and artifacts were formulated as follows: (1) (2) (3) (4) where EEG denotes the spectrum power summation of the . Because the electromyography (EMG) was bands not recorded in our system, EMG signals were not used in (1) and (4) [27]. The state index with the maximum value was extracted and the corresponding state, denoted as StageEEG, was the classication result of the epoch. In this study, the epochs were classied as wake (including wake and artifact) or sleep (including NREM and REM). 2) Result Reconrmation by ACC Analysis: After the state identication of an epoch was performed by EEG analysis, the result was reconrmed by ACC analysis. First, the magnitudes of the 3 acceleration channels were averaged as follows: (5) Then, the standard deviation of the averaged ACC signals in one epoch, dened as the movement power (MP), was calculated by the following: (6) Finally, as shown in Fig. 4 and (7), if an epoch was classied as a sleep state by the EEG and the MP of this epoch was larger than the higher threshold Thigh, the epoch was reclassied as a wake state. If an epoch was classied as a wake state and the , MP of this epoch was smaller than the lower threshold then it was reclassied as sleep. If the MP falls between the two thresholds, then the discrimination result based on the EEG was not changed

Fig. 4. Flowchart of the wakesleep state discrimination algorithm.

IV. SIGNAL ANALYSIS METHODS Two applications, automatic wakesleep state discrimination and temporal lobe epileptic seizure detection, are proposed in this paper, to demonstrate the advantages of the developed wireless physiological monitoring system, which integrates simultaneous EEG and accelerometer signal acquisitions and wireless transmission and is capable of performing long-term recording. Feature extraction and classication in the two applications have been described. Because rich information can be obtained from a combination of EEG and accelerometer signals, highly complicated signal analysis methods are not required for achieving excellent performance.

A. Identication of WakeSleep States A single channel of EEG recordings at the left-side frontal cortex (Cortex-L) and recordings of the three acceleration (ACC) channels (axis-X, axis-Y, and axis-Z) were combined to classify each 10-s epoch as either a wake or a sleep epoch [26]. Fig. 4 shows the owchart of the proposed wakesleep state identication method, in which the state of an epoch was rst determined based on the multiband EEG power spectra, and the result was then reconrmed according to the average magnitude of the three-axis ACC signals. For each 10-s epoch, the proposed method operated the following steps iteratively. 1) EEG Analysis: The EEG signals were calculated by a 1024-point fast Fourier transform (FFT). The resulting power spectra were divided into six bands, 0.51.5 Hz, 1.56 Hz, 69 Hz, 10.515 Hz, 2230 Hz, and 3545 Hz. By averaging its power magnitude of each sub-band, we further derive six arguments, EEGlo, , and . Afterwards, similar to [27],

(7) is the standard deviation of the avAssuming that eraged ACC signals in the training data (5-h data for rat I and and were calculated by the 4-h data for rat II), the following: (8) Fig. 5 shows the accuracy of the wakesleep state discrimination method with various values of and . According to

444

IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS, VOL. 1, NO. 4, DECEMBER 2011

Fig. 6. Distribution histogram of the values of normalized seizure power for nonseizure epochs (circles) and seizure epochs (crosses).

, the magnitudes of the ltered signals of three acceleration channels were rst averaged as

(9) We determined as the maximal energy, where were determined as follows: and (10) A moving average lter with a 61-epoch length was applied for smoothing, as follows: (11) The was used as the ACC feature for the th epoch, which was further normalized between 0 and 1 by dividing by the average SP values that corresponded to the seizure events in the training data (responses of two electrical stimulations), . The normalized feature was calculated by the following: (12) Finally, if was larger than a threshold , then this epoch was identied as an epoch in a clinical seizure; otherwise, it was regarded as a nonseizure epoch. Fig. 6 shows the distrifor seizure and nonseizure epochs in bution histogram of of the the training data of rat I. It can be observed that the seizure and nonseizure epochs can be successfully separated by was set at 0.8 in our experiment. a cut at 0.8. Therefore, V. SYSTEM EVALUATION The system that was developed was mounted on freely moving rats to perform various behavioral state and physiological signal monitoring experiments as a demonstration. Fig. 7

to
Fig. 5. Accuracy of the wakesleep state discrimination algorithm under different values of parameter g (a) and parameter g (b).

the analysis results, tively.

and

were set at 0.38 and 0.16, respec-

B. Automatic Temporal Lobe Epileptic Seizure Detection Conventionally, the clinical seizure onset was determined by the behavioral characteristics of the subject and the seizure was conrmed based on video examination by the experts, which was tedious and time-consuming. In addition, video recording required a large amount of storage space and was not suitable for automatic analysis. In this study, an automatic detection method of clinical seizures based on ACC signal analysis was developed. In our experiments, it was observed that the power spectrum of acceleration signals showed a strong power at 1020 Hz during a seizure. Therefore, ACC signals were ltered to extract the energy within 1020 Hz, and a thresholding approach was utilized for seizure detection. The recordings of the three acceleration channels were rst ltered by a fth-order band-pass Butterworth lter to extract the 1020 Hz components. For each nonoverlapping 1-s epoch

CHANG et al.: A VERSATILE WIRELESS PORTABLE MONITORING SYSTEM FOR BRAINBEHAVIOR APPROACHES

445

TABLE II DIMENSION, WEIGHT, AND CURRENT CONSUMPTION OF THE DATA ACQUISITION DEVICE

Fig. 7. The data acquisition device and battery pack installed on a rat. Placement of the three-axis accelerometer with respect to the rats head is shown in the top-right corner. The X, Y, and Z axes direct to the frontrear, updown, and leftright directions, respectively.

B. Board Dimension and Power Consumption The size and weight of the device are critical for a freely moving rat to carry it. Table II summarizes the dimensions and weight of the data acquisition device, which is powered by a 3.7 V, 1100 mAH, (HYB, China) Li-ion battery. The accelerometer board, mounted on the head of a rat, is 0.96 g, and the other boards together with the battery, mounted on a jacket worn by the rat, are 22.98 g. The battery life is also a crucial performance metric for a portable device. Table II presents the current consumption of the data acquisition device as well as the battery life. The current consumption was measured by using the National Instrument PXI-4071 7-1/2-digit Digital Multimeter. The battery operation was examined under a free running test until the system ran out of power; the time stamped testing log le indicated the battery life. As shown in Table II, the battery can support a 56-h operation.

Fig. 8. Timing diagram of the data acquisition and wireless transmission tasks.

shows a rat that is able to carry the data acquisition device without the hassles of wires and motion constraints. The axes of the accelerometer with respect to a rats head are illustrated in the top-right corner of Fig. 7. The X, Y, and Z axes correspond to the frontrear, updown, and leftright directions, respectively. One to four EEG channels at the frontal cortex and/or the amygdala were simultaneously recorded, depending on the different experimental settings. The video recording was also performed during the experiments, to conrm the behavioral states of the rats. A. Real-Time Monitoring and Data Transmission Fig. 8 presents the timing diagram of the data acquisition and wireless transmission tasks when all of the seven channels were used in an experiment. The timing was obtained by setting/resetting an MCU I/O pin at the start and end of each task and observing the duration of toggling of the I/O signal via an oscilloscope. As shown in Fig. 8, the transmission buffer became full every four sampling periods. Each time that the buffer became full, a 2.5 ms latency was required to transmit the buffered data via the wireless link. Although the wireless transmission spans into the next sampling period, the sampled data collection that is performed with higher priority is not interfered with.

VI. AUTOMATIC WAKESLEEP STATE DISCRIMINATION Discriminating the state of vigilance is a typical problem in a human or animal study [28], [29]. In this study, a single channel of EEG recordings at the left-side frontal cortex (Cortex-L) and recordings of the three acceleration (ACC) channels (axis-X, axis-Y, and axis-Z) were combined to classify each 10-s epoch as a wake or a sleep epoch by the proposed wakesleep state identication method. Long-term and uninterrupted EEG and ACC signals from two rats were recorded by the developed wireless and portable monitoring system. The lengths of the data from rat I and rat II were 10 h and 8 h, respectively. The uninterrupted recordings demonstrated that the developed system can successfully and continuously operate for more than 10 h. As shown in Table III, the data for rat I contain 2040 sleep epochs and 1560 wake epochs, and the data for rat II consist of 1420 sleep epochs and 1400 wake epochs. These results were obtained from the manual scorings of the EEG and ACC signals by an expert. Comparing the identication results of the proposed method and the expert shown in Table III, the accuracies for rats I and II are 84.44% and 89.40%, respectively. It is noted that the lack

446

IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS, VOL. 1, NO. 4, DECEMBER 2011

TABLE III CONFUSION MATRIX OF AUTOMATIC WAKESLEEP STATE DISCRIMINATION

TABLE IV CONFUSION MATRIX OF AUTOMATIC WAKESLEEP STATE DISCRIMINATION AFTER INDETERMINATE-EPOCH ELIMINATION

Fig. 9. (a) A 60-s recording of the EEG (Cortex-L). (b) Power spectrum of the 60-s recording of the EEG. (c) A 60-s recording of the acceleration signal (average values of the three ACC channels). (d) Movement power (MP) with and T (e) The 60-s results of the wakesleep state two thresholds T discrimination made by an expert, by EEG alone, and by a combination of EEG and ACC.

of EMG data could cause an expert to misclassify an epoch during the boundary of wake and sleep. By eliminating the indeterminate epochs that tend to cause confusion [7], as shown in Table IV, the accuracies for rats I and II were increased to 95.06% and 96.22%, respectively. The rejection rates, dened as the percentages of the indeterminate epochs, were 13.97% and 8.97% for rats I and II, respectively. Fig. 9 shows illustrative epochs for the demonstration of the effectiveness of combining EEG and acceleration signals for sleep-wake discrimination. Compared to the result based on EEG analysis alone, with reconrmation by ACC analysis, the misidentied epochs (15381540 and 15431544) can be successfully corrected without changing the states of the epochs that were correctly identied by the EEG analysis. The data of 10-h recordings from rat I and the results of applying the developed automatic wakesleep state identication method to these data were presented in Fig. 10. In addition to successful state identication, the stability of the system was also presented. According to the accelerometer signals, the rat made some large or rapid movements during the recording period, but these movements did not greatly interfere with the quality of the recorded EEG signals. This observation demonstrates the advantage of wireless transmission for high-quality recording and the stability of the developed system. Although various wireless physiological recording systems have been developed, reports demonstrating long-term continuous recording for more than 10 h are rare. The method that was developed can also be applied to automatically extract the wake epochs to benet the study of animal behavior. An example of behavior monitoring by analyzing the recorded signals during the identied wake epochs is shown in Fig. 11. Fig. 11(a) shows that the accelerometer signals at the x and y axes have a negative correlation during . This exploration. The correlation coefcient is about

Fig. 10. (a) A 10-h continuous recording of the EEG (Cortex-L). (b) A 10-h continuous recording of the acceleration signal (average values of the three ACC channels). (c) The 10-h results of the wakesleep state discrimination made by an expert, by EEG alone, and by a combination of EEG and ACC.

result shows synchronous motions of the head in frontrear and updown directions of body movements. Fig. 11(b) depicts noise signals that were caused by the fast motions of forelimbs when grooming. Signals during the identied sleep epochs are also shown for comparison, as in Fig. 11(c). During the sleep state, the accelerometer signals are very weak. This experiment demonstrates that combining the accelerometer with an EEG is an alternative to video-EEG monitoring, and the former provides sufcient information for automatic behavior analysis [7].

CHANG et al.: A VERSATILE WIRELESS PORTABLE MONITORING SYSTEM FOR BRAINBEHAVIOR APPROACHES

447

Fig. 12. Representative example of a temporal lobe epilepsy induced by amygdala kindling with EEGs of bilateral cortices and amygdala as well as three-axis acceleration signals and their summation. Racines stages are separated with bold solid lines.

Fig. 11. Behavior monitoring of the rat by using the developed system. (a) Exploration. (b) Grooming. (c) Sleep.

Fig. 13. Video frames (captured every 4 s) of a complete course of temporal lobe seizure behavior shown in Fig. 12.

VII. TEMPORAL LOBE EPILEPTIC SEIZURE DETECTION A fully kindled TLE, i.e., containing Racines stages 15, is shown in Fig. 12. Dynamic changes in the brain signals of the bilateral cortices and the amygdala were correlated with Racines stages and were conrmed with video examinations (Fig. 13). Based on a violent behavior during the TLE development, the wireless transmission is shown to have substantial benet for the convulsive TLE recording. According to the Racines seizure stage, a three-axis accelerator revealed various patterns during TLE development. For example, an x-axis accelerometer represented the rostrocaudal direction (Fig. 7), and it could clearly display a large amount of activity during the head movement or nodding that is related to Racines stage 2 of TLE (Fig. 12). To reduce computational complexity, the three-axis accelerometer signals were averaged. The averaged accelerometer signals used in the classication of the vigilance states also possessed comparable features during the TLE development. In this study, a rst important step is to develop a robust and useful system to assist TLE identication. Fig. 14 shows a comparison of the performance of TLE identication using EEG features or averaged accelerometer features in a 3.2-h recording of rat II. From the time-frequency spectra of either an EEG or an averaged accelerometer, high-frequency power was noticed during TLE development [30], [31]. Here, the power of 1020 Hz was selected as a feature to identify TLE through a xed threshold (Section IV). The threshold was empirically dened as the minimal power as the TLE occurred. The performance of the accelerometer feature was superior to that of the EEG feature [Fig. 14(e)]. The system performance was assessed in terms of accuracy and frequency of false alarms. A frequency of false alarms is dened as an error event number divided by an accurate event number. In the two rats, the accelerometer feature worked very well (100% accuracy and 0% false alarm frequency). However, the EEG feature had a high accuracy (100%) but revealed a very high frequency of false alarms (212.5% in rat I; 185.7% in rat II). The increased threshold in the EEG features resulted in an improvement in the false alarms in coincidence with a decreased accuracy. Based on these results, we demonstrated the advantage of the accelerometer for the TLE detection.

448

IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS, VOL. 1, NO. 4, DECEMBER 2011

Fig. 14. Representative examples of the identication of temporal lobe epilepsies. (a) Time-frequency power spectrum of left-side cortical EEG. (b) Summation of 1020 Hz power of the left-side cortical EEGs. Seizures detected by a xed threshold are marked by circles. (c) Time-frequency power spectrum of the three-axis averaged acceleration signal. (d) Summation of the 1020 Hz power of the three-axis averaged acceleration signal. Seizures that are detected by a xed threshold are marked by circles. (e) Correlations between expert and EEG and accelerator (ACC) systems. (S, seizure; NS, nonseizure).

VIII. DISCUSSION AND CONCLUSION The wireless physiological signal monitoring system was successfully developed and installed on freely moving rats for wakesleep classication, and TLE detection was performed through a combination of EEG and accelerometer. The system has the advantages of low cost, compact size, low power, and real-time monitoring. The combination of EEG and accelerometer sensors eliminates the need of a video camera for this application, simplifying the experimental setup. More importantly, the wireless communication frees experimental subjects from the hassle of wires and reduces wire artifacts of EEG signals during convulsions. The experiments on kindling processes in TLE, the discrimination of vigilance states, and long-run recordings were conducted, and they demonstrated that the EEG along with the accelerometer signicantly improves the physiological and behavioral observations. Discrimination of vigilance states is very important, and will help us to understand the functional performance of the brainbehavior relationship. However, this type of discrimination is a tedious and time-consuming task that requires a large amount of labor [3], [7]. Automatic classication systems have been proposed to solve the problem [7], [27]. In wakesleep classication, our proposed method has a slightly higher per% accuracy) than a previous study [7] under the formance ( condition of ruling out the indeterminate epoch, a strategy used in both studies. Our system provided the additional benet of being wireless. In the application of seizure detection, we provided strong evidence to support the use of an accelerometer for TLE detection in a continuous recording. According to the time-frequency spectrum of EEG, power uctuations were widespread through the entire frequency range without a consistent feature during

TLE development. It would be difcult to identify a suitable feature for accurate detection of seizures in a continuous recording with numerous artifacts [30]. Here, we provide an alternative way of attaining a high detection rate of TLE, i.e., incorporating an accelerometer. This system will be very helpful for reducing the large amount of manpower required previously for seizure detection. Current implementation of the proposed wireless physiological signal monitoring system can simultaneously record up to four channels of EEG signals and three channels of accelerometer signals. As shown in this paper, effective methods for automatic wakesleep classication and TLE detection were developed based on the proposed system, showing that the channels are sufcient for these studies. Moreover, the system has demonstrated the ability to monitor the behavior as well as full kindled TLEs of rats. The number of recording channels required in a study depends on the problem to be investigated. For example, online detection of absence seizure can be achieved by using a single EEG channel [32], [33]. However, all of the seven recording channels were used when monitoring a fully kindled TLE in this paper. Increasing the number of recording channels in the proposed system can extend its applicability. For example, more EEG channels help when classifying the Racine stages or when identifying the focus of an epileptic seizure. However, these channels require additional hardware components, such as the amplication circuits for the additional channels and a microcontroller with more ADC channels, which increases the systems complexity and shortens the duration of the operation. In addition, the bandwidth limitation of the wireless link should also be considered. ACKNOWLEDGMENT The authors would like to thank Texas Instruments for offering free wireless communication protocol stack and application programming interfaces which are the base of the proposed wireless recording system. REFERENCES
[1] E. Niedermeyer, , E. Niedermeyer and F. L. D. Silva, Eds., Sleep and EEG, in Electroencephalography Basic Principles, Clinical Applications and Related Fields, 4th ed. Baltimore, MD: Lippincott Williams Wilkins, 1999. [2] D. Neckelmann and R. Ursin, Sleep stages and EEG power spectrum in relation to acoustical stimulus arousal threshold in the rat, Psychopharmacology, vol. 16, no. 9, pp. 467477, 1993. [3] F. Z. Shaw, C. J. Lai, and T. H. Chiu, A low-noise exible integrated system for recording and analysis of multiple electrical signals during sleep-wake states in rats, J. Neurosci. Methods, vol. 118, no. 1, pp. 7787, 2002. [4] M. Sekine, M. Akay, T. Tamura, and Y. Higashi, Fractal dynamics of body motion in patients with Parkinsons disease, J. Neural Eng., vol. 1, no. 1, pp. 815, 2004. [5] P. Tass, D. Smirnov, A. Karavaev, U. Barnikol, T. Barnikol, I. Adamchic, C. Hauptmann, N. Pawelcyzk, M. Maarouf, V. Sturm, H. Freund, and B. Bezruchko, The causal relationship between subcortical local eld potential oscillations and Parkinsonian resting tremor, J. Neural Eng., vol. 7, no. 1, pp. 1600916024, 2010. [6] D. Rand, J. J. Eng, P. F. Tang, J. S. Jeng, and C. Hung, How active are people with stroke? Use of accelerometers to assess physical activity, Stroke, vol. 40, no. 1, pp. 163168, 2009. [7] S. Sunderama, N. Chernyya, N. Peixotob, J. P. Masona, S. L. Weinsteinc, S. J. Schiff, and B. J. Gluckman, Improved sleep-wake and behavior discrimination using MEMS accelerometers, J. Neurosci. Methods, vol. 163, no. 2, pp. 373383, 2007.

CHANG et al.: A VERSATILE WIRELESS PORTABLE MONITORING SYSTEM FOR BRAINBEHAVIOR APPROACHES

449

[8] L. D. Iasemidis, H. P. Zaveri, J. C. Sackellares, W. J. Williams, and T. W. Hood, Nonlinear dynamics of electrocorticographic data, J. Clin. Neurophysiol., vol. 5, p. 339, 1988. [9] L. D. Iasemidis, J. C. Sackellares, H. P. Zaveri, and W. J. Williams, Phase space topography and the Lyapunov exponent of electrocorticograms in partial seizures, Brain Topogr., vol. 2, no. 3, pp. 187201, 1990. [10] N. Kannathal, M. L. Choo, U. R. Acharya, and P. K. Sadasivan, Entropies for detection of epilepsy in EEG, Comput. Methods Programs Biomed., vol. 80, no. 3, pp. 187194, 2005. [11] A. Rosso, S. Blanco, J. Yordanova, V. Kolev, A. Figliola, M. Schrmann, and E. Basar, Wavelet entropy: A new tool for analysis of short duration brain electrical signals, J. Neurosci. Methods, vol. 105, no. 1, pp. 6575, 2001. [12] A. Rosso, Entropy changes in brain function, Int. J. Psychophysiol., vol. 64, no. 1, pp. 7580, 2007. [13] V. Srinivasan, C. Eswaran, and N. Sriraam, Approximate entropybased epileptic EEG detection using articial neural networks, IEEE Trans. Inf. Technol. Biomed., vol. 11, no. 3, pp. 288295, May 2007. [14] M. A. L. Nicolelis, Actions from thoughs, Nature, vol. 409, no. 6818, pp. 403407, 2001. [15] M. A. L. Nicolelis and M. A. Lebedev, Principles of neural ensemble physiology underlying the operation of brain-machine interfaces, Nat. Rev. Neurosci., vol. 10, no. 7, pp. 530540, 2009. [16] M. A. Castro-Alamancos, Vibrissa myoclonus (rhythmic retractions) driven by resonance of excitatory networks in motor cortex, J. Neurophysiol., vol. 96, no. 4, pp. 16911698, 2006. [17] A. K. Whitchurch, B. H. Ashok, R. V. Kumaar, K. Saurkesi, and V. K. Varadan, Wireless system for long-term EEG monitoring of absence epilepsy, in Proc. SPIE, 2002, pp. 343349. [18] C. T. Lin, L. W. Ko, J. C. Chiou, J. R. Duann, R. S. Huang, S. F. Liang, T. W. Chiu, and T. P. Jung, Noninvasive neural prostheses using mobile and wireless electroencephalography, Proc. IEEE, vol. 96, no. 7, pp. 11671183, Jul. 2008. [19] C. T. Lin, Y. C. Chen, T. Y. Huang, T. T. Chiu, L. W. Ko, S. F. Liang, H. Y. Hsieh, S. H. Hsu, and J. R. Duann, Development of wireless brain computer interface with embedded multi-task scheduling and its application on real-time drivers drowsiness detection and warning, IEEE Trans. Biomed. Eng., vol. 55, no. 5, pp. 15821951, May 2008. [20] S. L. Moshe, N. S. Sharpless, and J. Kaplan, Kindling in developing rats: Variability of after discharge thresholds with age, Brain Res., vol. 211, no. 1, pp. 190195, 1981. [21] J. Nissinen, T. Halonen, E. Koivisto, and A. Pitkanen, A new model of chronic temporal lobe epilepsy induced by electrical stimulation of the amygdala in rat, Epilepsy Res., vol. 38, no. 23, pp. 177205, 2000. [22] R. J. Racine, Modication of seizure activity by electrical stimulation. II: Motor seizure, Electroencephalogr. Clin. Neurophysiol., vol. 32, no. 3, pp. 281294, 1972. [23] Texas Instrument, CC2430 Software Examples Users Guide [Online]. Available: http://www.ti.com/litv/pdf/swru178 [24] Freescale Semiconductor, 1:5 g6g Three Axis Low-g Micromachined Accelerometer, Datasheet [Online]. Available: http://www. freescale.com/les/sensors/doc/data_sheet/MMA7260QT.pdf [25] Texas Instrument, A True System-on-Chip Solution for 2.4 GHz IEEE 802.15.4/ZigBee (Rev. F) [Online]. Available: http://www.ti.com/lit/ gpn/cc2430 [26] S. Sorge, T. Pollmcher, and M. Lancel, Clozapine alters sleep-wake behavior in rats, Neuropsychopharmacol., vol. 29, no. 8, pp. 14621469, 2004. [27] R. Stephenson, A. M. Caron, D. B. Cassel, and J. C. Kostela, Automated analysis of sleep-wake state in rats, J. Neurosci. Methods, vol. 184, no. 2, pp. 263274, 2009. [28] Y. T. Peng, C. Y. Lin, M. T. Sun, and C. A. Landis, Multimodality sensor system for long-term, sleep quality monitoring, IEEE Trans. Biomed. Circuits Syst., vol. 1, no. 3, pp. 217227, Sep. 2007. [29] W. Karlen, C. Mattiussi, and D. Floreano, Sleep and wake classication with ECG and respiratory effort signals, IEEE Trans. Biomed. Circuits Syst., vol. 3, no. 2, pp. 7178, Apr. 2009. [30] A. M. White, P. A. Williams, D. J. Ferraro, S. Clark, S. D. Kadamb, F. E. Dudek, and K. J. Staley, Efcient unsupervised algorithms for the detection of seizures in continuous EEG recordings from rats after brain injury, J. Neurosci. Methods, vol. 152, no. 12, pp. 255266, 2006. [31] G. Bettus, F. Wendling, M. Guye, L. Valton, J. Regis, P. Chauvel, and F. Bartolomei, Enhanced EEG functional connectivity in mesial temporal lobe epilepsy, Epiepsy Res., vol. 81, no. 1, pp. 5868, 2008.

[32] S. F. Liang, Y. C. Liao, F. Z. Shaw, D. W. Chang, C. P. Young, and H. Chiueh, Closed-loop seizure control on epileptic rat models, J. Neural Eng., vol. 8, no. 4, p. 45001, 2011. [33] C. P. Young, S. F. Liang, D. W. Chang, Y. C. Liao, F. Z. Shaw, and C. H. Hsieh, A portable wireless on-line closed-loop seizure controller in freely moving rats, IEEE Trans. Instrum. Meas., vol. 60, no. 2, pp. 513521, 2011. Da-Wei Chang (M08) received his B.S., M.S., and Ph.D. degrees in computer and information science from National Chiao Tung University, Hsinchu, Taiwan, in 1995, 1997, and 2001, respectively. He was a Postdoctoral Researcher with National Chiao Tung University in 20022005, and an Assistant Professor with the Department of Electrical Engineering, National Sun Yat-Sen University, Kaohsiung, Taiwan, in 2006. He is currently an Assistant Professor with the Department of Computer Science and Information Engineering, National Cheng Kung University, Tainan, Taiwan. His research interests include embedded systems for biomedical applications, wireless monitoring and control systems, and system software.

Sheng-Fu Liang (M09) received the B.S. and M.S. degrees in control engineering from the National Chiao Tung University (NCTU), Taiwan, in 1994 and 1996, respectively, and the Ph.D. degree in electrical and control engineering from NCTU, in 2000. From 2001 to 2005, he was a Research Assistant Professor in Electrical and Control Engineering, NCTU. He joined the Department of Biological Science and Technology, NCTU, in 2005 and joined the Department of Computer Science and Information Engineering (CSIE) and Institude of Meidical Informatics (IMI), National Cheng Kung University (NCKU), Tainan, Taiwan, in 2006. Currently, he is an Accociate Professor in CSIE and IMI, NCKU. He is also a collaborative researcher of Brain Research Center (BRC) and Biomimatic Systmes Research Center (BSRC), NCTU. His current research interests are neural engineering, biomedical engineering, biomedical signal/image processing, machine learning, and medical informatics.

Chung-Ping Young (S97M06) received the B.S. degree in electronic engineering from Chung Yuan Christian University, Taiwan, in 1985, and the M.S. and Ph.D. degrees in electrical engineering from the University of Missouri, Columbia, in 1994 and 1997, respectively. From 1994 to 1997, he was a Research Assistant at the University of Missouri, where he worked on power measurement. From 1998 to 2003, he was with Wistron InfoComm and Phoenix Technologies as a Senior Engineer and Principal Engineer. Since 2003, he has been with the Department of Computer Science and Information Engineering, National Cheng Kung University, Taiwan, where he is currently an Associate Professor. His elds of interest include vehicular telematics, multi-core embedded software, and distributed instrumentation.

Fu-Zen Shaw received the B.S. in biomedical engineering from Chung Yuan Christian University, Chungli, the M.S. and Ph.D. degrees in electrical engineering from National Taiwan University, Taipei, Taiwan. He worked at Tzu Chi University and National Chiao Tung University from 1999 to 2006. He is currently a Professor and the Head of the Department of Psychology, with a joint appointment in the Institute of Gerontology, National Cheng Kung University. His research interests include the

450

IEEE JOURNAL ON EMERGING AND SELECTED TOPICS IN CIRCUITS AND SYSTEMS, VOL. 1, NO. 4, DECEMBER 2011

establishing animal models of neurological diseases (epilepsy, bromyogia, stroke, sleep disorders) for mechanism investigation and therapeutic platform, development of neuro-cognitive rehabilitation, neurofeedback training, sleep and memory, and home-based health research.

Alvin W. Y. Su received the B.S. degrees in control engineering from National Chiao-Tung University (NCTU), Taiwan, in 1986. He received the M.S. and Ph.D. degrees in electrical engineering from Polytechnic University, Brooklyn, NY, in 1990 and 1993, respectively. From 1993 to 1994, he was with Center for Computer Research in Music and Acoustics (CCRMA), Stanford University, Stanford, CA. From 1994 to 1995, he was with Computer Communication Lab of the Industrial Technology Research Institute (CCL. ITRI.), Taiwan. In 1995, he joined the Department of Information Engineering and Computer Engineering at Chung-Hwa University, where he serves as an Associate Professor. In 2001, he joined the Department of Computer Science and Information Engineering, National Cheng-Kung University. His research interests include digital audio signal processing, physical modeling of acoustic musical instruments, human computer interface design, video and color image signal processing, and VLSI signal processing.

Yi-Che Liu received the B.S. degree in computer science and information engineering from National Cheng Kung University, Taiwan, in 2008. He is currently working toward the M.S. degree with the Department of Computer Science and Information Engineering, National Cheng Kung University, Taiwan. His research interests include biomedical signal processing and sleep disorder analysis.

Jing-Jhong Chen received the B.S. degree in industrial education from National Taiwan Normal University, Taiwan, in 2008. He is currently working toward the M.S. degree with the Department of Computer Science and Information Engineering, National Cheng Kung University, Taiwan. His research interests include braincomputer interface and electronic circuits design.

You-De Liu received his B.S. and M.S. degrees in computer science and information engineering from National Cheng Kung University, Taiwan, in 2006 and 2008, respectively. He is currently working toward the Ph.D. degree with the Department of Computer Science and Information Engineering, National Cheng Kung University, Taiwan. His research interests include embedded systems design, biomedical signal processing, and wireless communication.

Chun-Yu Chen received the B.S. degree in computer science and information engineering from National Cheng Kung University, Taiwan, in 2008. He is currently working toward the M.S. degree with the Institute of Medical Informatics, National Cheng Kung University, Taiwan. His research interests include biomedical signal analysis and brain-computer interface design.

Yu-Lin Wang received his B.S. and M.S. degrees in computer science and information engineering from National Cheng Kung University, Taiwan, in 2005 and 2007, respectively. She is currently working toward the Ph.D. degree with the Department of Computer Science and Information Engineering, National Cheng Kung University, Taiwan. Her research interests include embedded systems design, biomedical signal processing, and data compression.

Das könnte Ihnen auch gefallen